Despite the rapid evolution of ultrafast lasers over the past decades, the need for ever higher pulse energies and peak and average powers remains undiminished. In particular, most demanding technologies, such as laser particle acceleration [1] and high-harmonic generation [2], continue driving the development of promising performance scaling techniques. Among these, the various implementations of spatial [3–6] and temporal coherent pulse combination [7–9] are valuable examples. They allow to exceed the limitations of a single pulse in a single amplifier by parallelization, using a set of parallel amplifiers, a burst of temporally separated pulses, or a combination of both instead. Due to a good and reproducible beam quality and a typically barely exhausted energy extraction, both techniques are particularly effective when used with fiber amplifiers, regularly enabling record-breaking results [5,6,9].

In this contribution, we present a high-energy, high-power ultrafast fiber laser system employing spatiotemporal coherent combination of 128 pulse replicas by operating 16 separate ytterbium-doped fiber amplifiers with bursts of eight consecutive pulses each. The chirped-pulse amplification (CPA) main amplifier stage and the spatial coherent combination are based on the system presented in [6]. The additionally implemented temporal combining is a modified version of electro-optically controlled divided-pulse amplification (EDPA) [9,10] using intensity combining. Furthermore, all three required optical delay lines are integrated in a single, ultra-compact imaging multipass cell (MPC). Finally, almost all energy stored in the 16 amplifier fibers is extracted and previously achieved record results [9] are surpassed anew.

A schematic illustration of the setup is depicted in Fig. 1. The completely fiber-integrated and polarization-maintaining front end is seeded with an ytterbium-fiber-based broadband oscillator, which generates femtosecond pulses at a repetition rate of $f_\mathrm {rep}={80}\,\textrm{MHz}$. A first set of chirped fiber Bragg gratings (CFBGs) with a spectral hard cut of 1015 nm to 1055 nm stretches the pulses to the nanosecond range. Next, the signal passes a Fourier-domain pulse shaper (FDPS), which allows manipulating the spectral phase and amplitude to compensate gain narrowing and residual higher order dispersion. A first acousto-optic modulator (AOM 1) picks bursts of eight subsequent pulses from the signal pulse train at a burst repetition rate of 400 kHz. Thus, the distance between two subsequent bursts is 2.5 µs, while the separation of two adjacent pulses within a burst is $\tau =1/f_\mathrm {rep}={12.5}\,\textrm{ns}$. A second set of CFBGs completes the stretcher dispersion to a total of $\beta _2\,{=}\,{127}\,\textrm{ps}^2$ and $\beta _3\,{=}\,{-2.7}\,\textrm{ps}^3$, corresponding to a stretched pulse duration of more than 10 ns if the spectral hardcut is filled. The following AOM 2 reduces the burst repetition rate to a final value of 20 kHz. In addition, as it is controlled by an arbitrary waveform generator (AWG), it allows free amplitude shaping of the individual pulse replicas within the burst. This is essential to pre-compensate gain-saturation-induced burst deformation effects, which inevitably arise especially in the main amplifiers as significant energy extraction is pursued. Therefore, to maximize the extracted energy and temporal combining efficiency, amplitude pre-shaping is key [11]. Next, the signal is split into two equal fiber channels, each containing an electro-optic modulator (EOM 1 and EOM 2) driven by a two-channel AWG, which allows to individually apply distinctive phase patterns on both pulse bursts. Finally, these are coupled out of the fiber integrated front end as two separate beams. In between, the signal is repeatedly re-amplified in the fiber front end by core-pumped ytterbium-doped amplifiers to compensate signal losses in fiber components and pulse picking.

 

Fig. 1. Schematic of the setup. The pulses on the left side of the intensity splitters in the temporal combining stage are sent into the respective optical delay lines (DL1, light green; DL2, medium green; DL3, dark green), while the pulses on the right receive no additional delay. CFBGs, chirped fiber Bragg gratings; FDPS, Fourier-domain pulse shaper; AOM, acousto-optic modulator; EOM, electro-optic modulator; MDC, multi delay line cell.

Download Full Size | PDF

The following pre- and main amplifier stages are arranged as two equal but independent layers on top of each other with each layer being seeded by one of the two beams emitted by the fiber front end. Thus, the beam path is described for only one of them and it is, apart from the preamplifier stage being parallelized, mostly similar to the one described in [6].

The seed beam coupled out of one EOM fiber channel passes two free-space spectral notch filters used to further compensate gain narrowing and increase the finally available bandwidth. The signal is amplified in a first large mode area (LMA) fiber amplifier counter-pumped at 976 nm wavelength and further spectrally shaped by four additional free-space filters. Finally, the average power is raised to 12 W in a second LMA preamplifier. The beam is split by a set of polarizers and half-wave plates (HWPs) into eight separate channels to seed the eight parallel main amplifiers per layer. Each of them consists of a 105 cm-long ytterbium-doped LMA fiber with 62 µm mode field diameter, placed in a water-cooled aluminum module and counter-pumped separately at 976 nm wavelength with an individually settable power of up to 250 W. The eight amplified beams are spatially combined using thin-film polarizers (TFPs) and HWPs. The required interferometric phase stabilization is based on Hänsch–Couillaud polarization detection [12] and piezo-mounted mirrors on the seed side between the main amplifiers. In conclusion, there are now two output beams (one per layer) with a highly amplified eight-pulse burst each, differing only by the phase pattern applied earlier with the EOMs.

Both beams are collimated to a diameter of 4 mm and sent into the temporal combining stage. Its functioning principle is analogous to [9,10], using free-space optical delay lines and combining elements. However, few-degree angle of incidence (AOI) dielectric intensity beam splitters (BS) are now used instead of TFPs, as they possess superior thermal properties [13].

In the temporal combining, the phase patterns become key. In the first step, where both bursts are combined in a BS with 50% reflectivity, the phase differences between them are relevant. As this difference is alternating between $0$ and $\pi$ from pulse to pulse, the side of the BS where the pulses interfere constructively alternates accordingly. Thus, odd- and even-numbered pulse replicas become spatially separated. This allows to send pulses 1, 3, 5, and 7 into an additional optical delay, matching the distance to their respective even-numbered followers, pulses 2, 4, 6, and 8. Correspondingly, the delay length is $L_\mathrm {DL1}\,{=}\,\tau c\,{=}\,{3.75}\,\textrm{m}$, with the speed of light in air $c$. Thus, reuniting the spatially separated signals thereafter in a second BS results in a pairwise interferometric superposition of the formerly temporally separated pulse replicas 1 and 2, 3 and 4, 5 and 6, and 7 and 8. Here, again, the phase difference between both these incoming signals alternates between $0$ and $\pi$, as determined by the initial EOM phase pattern. At this point, the scheme repeats itself, with the pulse-to-pulse separation doubling and the number of pulses halving after each combination step. Thus, the delay line in this second step has twice the length $L_\mathrm {DL2}\,{=}\,2\tau c\,{=}\,{7.5}\,\textrm{m}$, and a following third step incorporates $L_\mathrm {DL3}\,{=}\,4\tau c\,{=}\,{15}\,\textrm{m}$. This principle is arbitrarily scalable, using bursts of $2^N$ pulses, $N$ delay lines with doubling lengths, and corresponding phase patterns on both input bursts. Therefore, the eight-pulse bursts are stacked into a single pulse again after three combining steps. An alternative approach to understanding and finding the appropriate input phase patterns can be found by theoretically propagating a single pulse backward through the temporal combining stage.

Because natural divergence and small pointing fluctuations of the input beam can have a severe impact on the combining efficiency and stability due to the long lengths of the delay lines, these are desired to have imaging properties. While Herriott-type multipass cells (MPCs) [14] can fulfill these demands, the typically oscillating beam spot sizes on the MPC mirrors can lead to potentially harmful fluences. Furthermore, requiring a complete MPC for each delay line makes scaling the pulse count less attractive. Thus, a new design was developed, integrating all delay lines in a single ultra-compact module with constantly large beam diameters on the optics. This multi delay line cell (MDC) is based on a detuned double-confocal resonator [15] and consists of four concave mirrors facing each other pairwise in a confocal distance, as depicted in Fig. 1. Consequently, the setup is comparable to a Herriott MPC in confocal configuration but with both mirrors sawed vertically in halves. A full round trip is achieved after passing each mirror once and, therefore, completes the 8$f$ imaging condition. By slightly tilting two of the mirrors, a constant vertical parallel offset of the beam is achieved after each full round trip instead of performing a closed beam path loop. Thus, by placing a small outcoupling mirror at the corresponding position and height, the beam path within the MDC can be intercepted after a desired number of round trips. Using MDC mirrors with a radius of curvature and distance of 937 mm, a full round trip amounts to 3.75 m. Thus, the three optical delay lines can be integrated in the MDC by using three separate pairs of in- and outcoupling mirrors, placed so that they generate 1, 2, and 4 full round trips, respectively. Furthermore, as a single round trip already fulfills the imaging condition, so do multiples of it and all beam spots on the mirrors have the same diameter as the collimated input and output beams, i.e., 4 mm. The foci are in the middle of the cell in free space.

Similar to the spatial combining of the parallelized main amplifiers, interferometric phase stabilization is required in the temporal combining. Thus, again, piezo-mounted mirrors are implemented in each delay line step and on the seed side between the two main amplifier layers, allowing to apply small rapid phase corrections. The corresponding error signal detection is based on LOCSET [16] using dither frequencies of ${3.5}\,\textrm{kHz}$, ${4.6}\,\textrm{kHz}$, ${6.6}\,\textrm{kHz}$, and ${7.5}\,\textrm{kHz}$. For this, a sample is taken from the combined signal and sent through an AOM on a photodiode. The AOM is implemented for temporal gating of the error signal, which is required to avoid incorrect phase-locking states [17]. In the future, motorized linear stages will be added to increase the phase correction range and counteract long-term drifts.

The finally completely combined single pulse in a single beam is then sent into a two-staged CPA grating compressor, which is partially filled with a protective gas atmosphere to avoid nonlinear absorption in ambient air [6]. Last, a sample of the compressed output beam is taken for analysis and measurements.

Before starting the experiment, the seed signal has to be further prepared. The seed bursts require a ramp shape with increasing amplitude, because the inversion in the following main amplifiers is depleted by each pulse and, consequently, the gain is reduced. Thus, appropriate preshaping ensures the energy extracted from the main amplifiers is well distributed over the entire burst and keeps the first pulse of it far from intensities potentially harmful to the amplifier or the pulse quality. Furthermore, the nonlinear phase accumulated by the individual pulse replicas has to be matched, as this is most important to allow a high combining efficiency [18]. As described in [9], EDPA allows to pick individual pulses for analysis by applying an appropriate phase pattern via the EOMs without altering the amplification characteristics. This is first used to minimize the spectral phase of the first pulse using the multiphoton intrapulse interference phase scan (MIIPS) technique [19], the FDPS in the fiber front end, and one main amplifier per layer. Next, the following pulses are investigated individually with an auto-correlator after CPA compression. The amplitude shape applied by AWG-driven AOM 2 is optimized until all pulse replicas have a similar compressed pulse shape and duration, which proved to be an effective indicator for the accumulated nonlinear phase. The resulting shapes of a burst before and after a single main amplifier are depicted in Fig. 2, and correspond to a contained energy of approximately 75 µJ and 3.3 mJ, respectively. The different envelope shapes of bursts before and after amplification are used to calculate the extractable energy by fitting the Frantz–Nodvik model [20] to them. Consequently, approximately 70% of the extractable energy is extracted from the employed amplifier fibers in this operational regime. Next, the remaining amplifier channels are matched for optimum combining efficiency using the individually settable pump powers. After spatially combining all eight amplifiers per layer, the resulting two beams have an average power of 964 W together, corresponding to an available burst energy of 48 mJ.

 

Fig. 2. Shapes of the burst before (red) and after (blue) main amplification, resulting in matching B-integrals of all pulse replicas. Measured with a photodiode with 12 ps rise time and an oscilloscope with a sampling rate of 80 GS/s.

Download Full Size | PDF

These two beams are sent into the temporal combining stage, resulting in a combined signal with 825 W average power. Dividing it by the available power before yields a combining efficiency of the temporal combining stage of $\eta _\mathrm {comb}\,{=}\,86\%$, whereby most of the missing power can be associated with the last combining step. Losses caused here by an energetic mismatch of the two incoming final pulses, which remain after combining the first four and the last four pulses respectively in the preceding steps, have been minimized by using a final BS with 65% reflectivity. However, small losses remain, for instance, due to spectral or phase front mismatches or imperfect adjustment, manifesting as power losses dumped in the last step, and pre and post pulses in the combined beam. The latter is investigated by temporally resolving the combined signal with a fast photodiode as depicted in Fig. 3. Analyzing this trace reveals that 90% of the total energy contained in the depicted signal belongs to the main pulse while, consequently, only 10% are residual pre and post pulses. The former value is referred to as the temporal efficiency $\eta _\mathrm {temp}$ and, together with the combining efficiency, allows to summarize the performance of the temporal combining stage as the stacking efficiency $\eta _\mathrm {stack}\,{=}\,\eta _\mathrm {comb}\eta _\mathrm {temp}\,{=}\,77\%$. It is a comprehensive figure of merit for the performance of temporal combining and represents the obtained actual pulse energy after combination in relation to the total energy available before. Accordingly, the combined main pulse energy in this experiment is 37 mJ. Furthermore, the pulse contrast is determined, defined here as the energy ratio of the main pulse over the most intense side pulse in Fig. 3, resulting in $C\,{=}\,{18}\,\textrm{dB}$. While this is a good result for temporal combination and suitable for many purposes as, for instance, high-harmonic generation [2], more sensitive applications would require additional contrast enhancement coming at the expense of average power and pulse energy [21,22]. Next, the optical spectrum of the combined signal is measured, showing a ${20}\,\textrm{dB}$ bandwidth of more than 24 nm in Fig. 4.

 

Fig. 3. Spatiotemporally combined output signal, measured with a 12 ps rise time photodiode and an 80 GS/s oscilloscope.

Download Full Size | PDF

 

Fig. 4. Spectrum of the spatiotemporally combined signal measured with a resolution of 100 pm.

Download Full Size | PDF

The combined signal is now recollimated in a lens telescope to a beam diameter of 8 mm and sent into the two-staged grating compressor which has a total transmissivity of 85%. Thereafter, an average power of 703 W is available which, using the previously determined temporal efficiency, implies a finally combined and compressed main pulse energy of 32 mJ. The compressed pulse is characterized with an auto-correlator and compared with the theoretical Fourier transform limit of the previously measured spectrum. Especially due to the low nonlinearity, enabled by the vast spatiotemporal energy distribution and an estimated B-integral of ${2}\,\textrm{rad}$ in the main amplifiers, a basically perfect pulse compression is achieved, depicted in Fig. 5. Using the deconvolution factor calculated from the spectrum, a pulse duration of 158 fs is obtained, both for the measured pulse and the Fourier transform limit.

 

Fig. 5. Auto-correlation measurement of the compressed pulse in blue and the Fourier transform limit (FTL) in red.

Download Full Size | PDF

Last, the beam quality of the combined and compressed signal is investigated. With $M^2\,{=}\,1.2$ in both axes, which is similar to the performance of the laser system without EDPA, no notable negative impact of the temporal pulse stacking or the MDC is observed.

In conclusion, implementing the pulse energy scaling technique EDPA into an existing 16-channel fiber laser system running at a pulse repetition rate of 20 kHz allowed to more than triple its compressed pulse energy to 32 mJ, the highest value ever reported for a fiber CPA laser system. A good overall stacking efficiency of 77% is achieved, representing the percentage of the energy launched by the laser as an eight-pulse burst that is successfully stacked into a single pulse. Excellent beam and pulse quality after CPA compression are achieved with an $M^2\,{=}\,1.2$ and a pulse duration of 158 fs, allowing for a peak power of approximately 190 GW. In contrast to previous experiments, no performance loss is observed at highest average powers, and fluences on delay line mirrors are far from being critical. This is owed to a change from polarization to intensity beam combining and a newly designed ultra-compact multipass cell. Thus, EDPA is expected to play an important role in further pulse energy scaling of ultrafast laser systems, especially fiber-based ones, and help to reach the requirements of most demanding applications.